Hi, I just came back to giving the fastai another chance (or rather the motivation and some time) and started watching the first lessons (2022) and reproducing some of the notebooks. However when coding and learning I found that my progress seems to be quite impaired by the fact that there is no real thorough documentation of all the functions and classes! I’m quite annoyed by that. I’m looking for attributes, methods of classes, functions etc and don’t find them in the docs. Even with some digging and using e.g. ?function it doesn’t all really become clear to me.
I searched the forum and found a couple of “documentation” threads but it seems like the efforts have halted? E.g. in documentation-improvements the last post is from 2020 and many links are dead!
How am I supposed to learn fastai like that? Have I overlooked something? I see myself spending HOURS looking for explanations for the classes, methods etc. That is not exactly “fast” …
Generally you want to do show_doc(x) to see the full docments included (parameter documentation), as sadly they’re not attached to the docstrings.
The fastai docs are far from perfect, but generally I recommend viewing it on the full documentation website or in github itself and use searching.
Is there an easier way to use show_doc than importing it via from nbdev import show_doc every time?
I was gonna look at the link (https://docs.fast.ai/gen_doc.gen_notebooks.html) from the documentation-improvements thread but it’s broken.
E.g. Recorder. How am I supposed to find some good info on this class? show_doc gives me a one-liner. In the “full documentation website” as you call it, there barely is any info. When I dir(Recorder) it gives me lots of more methods/attributes that aren’t even mentioned in the “full documentation”.
It seems like the documentation assumes that many things are self-explanatory, implied or they are already explained elsewhere. Well, it would be nice to have references at the objects that point me to where it is already explained then.
Now I can show_doc into methods/attributes of e.g. Recorder but doing this for each attribute/method (for several objects) is very tedious. Similarly with searching github. Do you see my point?
I thought fastai is supposed to make things easier!
Nope, that’s the way. I don’t disagree with your points at all, and they are quite valid. fastai and how it’s programmed makes this extremely hard. The usual answer given is to explore the notebooks that they were built in, as then you can see the tests to try and help there. But I agree it is very frictionful.